| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
On some benchmarks provided by Chantal Keller a long time ago,
romega was abnormally slow compared to omega (or lia).
It turned out that the change of concl by reified version was
triggering unnecessary unfolds of Z.add, instead of unfolding
ReflOmegaCore.Z_as_Int.plus into Z.add. This is now fixed by
the various Parameter Inline : no more indirections, Z_as_Int.plus
is directly Z.add.
Also use Tactics.convert_concl_no_check for this "change", as
recommended by PMP.
|
|
|
|
|
|
|
|
|
|
Reminder of (some of) the reasons for removal:
- Despite the claim in sigma.mli, it does *not* prevent evar
leaks, something like:
fun env evd ->
let (evd',ev) = new_evar env evd in
(evd,ev)
will typecheck even with Sigma-like type annotations (with a proof of
reflexivity)
- The API stayed embryonic. Even typing functions were not ported to
Sigma.
- Some unsafe combinators (Unsafe.tclEVARS) were replaced with slightly
less unsafe ones (e.g. s_enter), but those ones were not marked unsafe
at all (despite still being so).
- There was no good story for higher order functions manipulating evar
maps. Without higher order, one can most of the time get away with
reusing the same name for the updated evar map.
- Most of the code doing complex things with evar maps was using unsafe
casts to sigma. This code should be fixed, but this is an orthogonal
issue.
Of course, this was showing a nice and elegant use of GADTs, but the
cost/benefit ratio in practice did not seem good.
|
|
|
|
This is the continuation of #244, we now deprecate `CErrors.error`,
the single entry point in Coq is `user_err`.
The rationale is to allow for easier grepping, and to ease a future
cleanup of error messages. In particular, we would like to
systematically classify all error messages raised by Coq and be sure
they are properly documented.
We restore the two functions removed in #244 to improve compatibility,
but mark them deprecated.
|
|
We move Coqlib to library in preparation for the late binding of
Gallina-level references. Placing `Coqlib` in `library/` is convenient
as some components such as pretyping need to depend on it.
By moving we lose the ability to locate references by syntactic
abbreviations, but IMHO it makes to require ML code to refer to
a true constant instead of an abbreviation/notation.
Unfortunately this change means that we break the `Coqlib`
API (providing a compatibility function is not possible), however we
do so for a good reason.
The main changes are:
- move `Coqlib` to `library/`.
- remove reference -> term from `Coqlib`. In particular, clients will
have different needs with regards to universes/evar_maps, so we
force them to call the (not very safe) `Universes.constr_of_global`
explicitly so the users are marked.
- move late binding of impossible case from `Termops` to
`pretying/Evarconv`. Remove hook.
- `Coqlib.find_reference` doesn't support syntactic abbreviations
anymore.
- remove duplication of `Coqlib` code in `Program`.
- remove duplication of `Coqlib` code in `Ltac.Rewrite`.
- A special note about bug 5066 and commit 6e87877 . This case
illustrates the danger of duplication in the code base; the solution
chosen there was to transform the not-found anomaly into an error
message, however the general policy was far from clear. The long
term solution is indeed make `find_reference` emit `Not_found` and
let the client handle the error maybe non-fatally. (so they can test
for constants.
|
|
We remove redundant functions `coq_constant`, `gen_reference`, and
`gen_constant`.
This is a first step towards a lazy binding of libraries references.
We have also chosen to untangle `constr` from `Coqlib`, as how to
instantiate the reference (in particular wrt universes) is a
client-side issue. (The client may want to provide an `evar_map` ?)
c.f. #186
|
|
The trace only mentions the constant k by which we want to divide
the equation, not anymore the equation we obtain after the division.
Shorter trace, and it won't take much more time to perform the few
Z.div than checking as currently the equality of the initial equation
and the final equation multiplied by k.
|
|
|
|
|
|
|
|
We benefit from the fact that we normalize now *all* hypotheses
even the one defining the "stated" variable: it is produced as
...def of v... = v
and normalized as
-v + ...def of v... = 0
which is precisely what we should add to the initial equation during
a O_STATE.
|
|
Now that O_SUM is properly optimized (cf. the [fusion] function),
we could use it to encode CONTRADICTION and NEGATE_CONTRADICT(_INV).
This way, the trace has almost the same size, but ReflOmegaCore.v
is shorter and easier to maintain.
|
|
|
|
|
|
|
|
In this variant, the proof term produced by romega isn't verified at
the tactic run-time (no vm_compute). In theory, [unsafe_romega] should
behave exactly as [romega], but faster. Now, if there's a bug in romega,
we'll be notified only at the following Qed. This could be interesting
for debugging purpose : you could inspect the produced buggish term
via a Show Proof.
|
|
This is a major change :
- Generated proofs are quite shorter, since only the resolution trace remains.
- No time penalty mesured (it even tends to be slightly faster this way).
- Less infrastructure in ReflOmegaCore and refl_omega.
- Warning: the normalization functions in ML and in Coq should be kept
in sync, as well as the variable order.
- Btw: get rid of ML constructor Oufo
|
|
|
|
In a coming commit, we'll normalize terms by a Coq function
that will compare Tvar's instead of blindly applying a trace,
so let's speed-up these comparisons.
|
|
|
|
|
|
|
|
|
|
|
|
For the bug, see new test test_romega10 in test-suite/success/ROmega0.v.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This is cumbersome, because now code may fail at link time if it's not
referring to the correct module name. Therefore, one has to add corresponding
open statements a the top of every file depending on a Ltac module. This
includes seemingly unrelated files that use EXTEND statements.
|
|
|
|
|
|
|
|
|
|
There was no reason to keep them separate since quite a long time. Historically,
they were making Genarg depend or not on upper strata of the code, but since
it was moved to lib/ this is not justified anymore.
|
|
|
|
module)
For the moment, there is an Error module in compilers-lib/ocamlbytecomp.cm(x)a
|
|
For now, the pack name reuse the previous .cma name of the plugin,
(extraction_plugin, etc).
The earlier .mllib files in plugins are now named .mlpack.
They are also handled by bin/ocamllibdep, just as .mllib.
We've slightly modified ocamllibdep to help setting the -for-pack
options: in *.mlpack.d files, there are some extra variables such as
foo/bar_FORPACK := -for-pack Baz
when foo/bar.ml is mentioned in baz.mlpack.
When a plugin is calling a function from another plugin, the name
need to be qualified (Foo_plugin.Bar.baz instead of Bar.baz).
Btw, we discard the generated files plugins/*/*_mod.ml, they are
obsolete now, replaced by DECLARE PLUGIN.
Nota: there's a potential problem in the micromega directory,
some .ml files are linked both in micromega_plugin and in csdpcert.
And we now compile these files with a -for-pack, even if they are
not packed in the case of csdpcert. In practice, csdpcert seems
to work well, but we should verify with OCaml experts.
|
|
Note that this breaks the compatibility, in a beneficial way I believe. Tactics
defined in strict mode (i.e. through Ltac foo := ...) may not do an introduction
on a local identifier anymore. They must use the "fresh" primitive instead.
|
|
This patch splits pretty printing representation from IO operations.
- `Pp` is kept in charge of the abstract pretty printing representation.
- The `Feedback` module provides interface for doing printing IO.
The patch continues work initiated for 8.5 and has the following effects:
- The following functions in `Pp`: `pp`, `ppnl`, `pperr`, `pperrnl`,
`pperr_flush`, `pp_flush`, `flush_all`, `msg`, `msgnl`, `msgerr`,
`msgerrnl`, `message` are removed. `Feedback.msg_*` functions must be
used instead.
- Feedback provides different backends to handle output, currently,
`stdout`, `emacs` and CoqIDE backends are provided.
- Clients cannot specify flush policy anymore, thus `pp_flush` et al are
gone.
- `Feedback.feedback` takes an `edit_or_state_id` instead of the old
mix.
Lightly tested: Test-suite passes, Proof General and CoqIDE seem to work.
|
|
|
|
|
